mirror of
https://gitlab.com/scemama/qp_plugins_scemama.git
synced 2024-12-23 04:43:38 +01:00
124 lines
5.5 KiB
Python
124 lines
5.5 KiB
Python
#!/usr/bin/env python3
|
|
# !!!
|
|
import os, sys
|
|
# !!!
|
|
#QP_PATH=os.environ["QMCCHEM_PATH"]
|
|
#sys.path.insert(0,QMCCHEM_PATH+"/EZFIO/Python/")
|
|
# !!!
|
|
from ezfio import ezfio
|
|
from datetime import datetime
|
|
import numpy as np
|
|
from scipy.sparse.linalg import svds
|
|
from R3SVD_LiYu import R3SVD_LiYu
|
|
from RSVD import powit_RSVD
|
|
from R3SVD_AMMAR import R3SVD_AMMAR
|
|
import time
|
|
# !!!
|
|
fmt = '%5d' + 2 * ' %e'
|
|
# !!!
|
|
if __name__ == "__main__":
|
|
# !!!
|
|
if len(sys.argv) != 2:
|
|
print("Usage: %s <EZFIO_DIRECTORY>"%sys.argv[0])
|
|
sys.exit(1)
|
|
filename = sys.argv[1]
|
|
ezfio.set_file(filename)
|
|
# !!!
|
|
N_det = ezfio.get_spindeterminants_n_det()
|
|
A_rows = np.array(ezfio.get_spindeterminants_psi_coef_matrix_rows())
|
|
A_cols = np.array(ezfio.get_spindeterminants_psi_coef_matrix_columns())
|
|
A_vals = np.array(ezfio.get_spindeterminants_psi_coef_matrix_values())
|
|
nrows, ncols = ezfio.get_spindeterminants_n_det_alpha(), ezfio.get_spindeterminants_n_det_beta()
|
|
Y = np.zeros( (nrows, ncols) )
|
|
for k in range(N_det):
|
|
i = A_rows[k] - 1
|
|
j = A_cols[k] - 1
|
|
Y[i,j] = A_vals[0][k]
|
|
print("# # # # # # # # # # # # # # # # # # # # # #")
|
|
print('matrix dimensions = {} x {}'.format(nrows, ncols))
|
|
print("# # # # # # # # # # # # # # # # # # # # # # \n")
|
|
normY = np.linalg.norm(Y, ord='fro')
|
|
print( normY )
|
|
# !!!
|
|
print('Full SVD:')
|
|
t_beg = time.time()
|
|
U, S_FSVD, VT = np.linalg.svd(Y, full_matrices=0)
|
|
t_end = time.time()
|
|
rank = S_FSVD.shape[0]
|
|
energy = np.sum(np.square(S_FSVD)) / normY**2
|
|
err_SVD = 100. * np.linalg.norm(Y - np.dot(U,np.dot(np.diag(S_FSVD),VT)), ord='fro') / normY
|
|
print('rank = {}, energy = {}, error = {}%, CPU time = {} \n'.format(rank, energy, err_SVD, t_end-t_beg))
|
|
# !!!
|
|
np.savetxt('results_python/h2o_pseudo/SingValues_FullSVD.txt', np.transpose([ np.array(range(rank))+1, S_FSVD ]), fmt='%5d' + ' %e', delimiter=' ')
|
|
# !!!
|
|
t = 50
|
|
delta_t = 10
|
|
npow = 15
|
|
err_thr = 1e-3
|
|
maxit = 10
|
|
# !!!
|
|
print('RRR SVD Li & Yu:')
|
|
t_beg = time.time()
|
|
U, S_R3SVD, VT = R3SVD_LiYu(Y, t, delta_t, npow, err_thr, maxit)
|
|
t_end = time.time()
|
|
rank = S_R3SVD.shape[0]
|
|
energy = np.sum( np.square(S_R3SVD) ) / normY**2
|
|
err_SVD = 100. * np.linalg.norm(Y - np.dot(U,np.dot(np.diag(S_R3SVD),VT)), ord='fro') / normY
|
|
print('nb Pow It = {}, rank = {}, energy = {}, error = {} %, CPU time = {}\n'.format(npow, rank, energy, err_SVD, t_end-t_beg))
|
|
# !!!
|
|
err_R3SVD = np.zeros( (rank) )
|
|
for i in range(rank):
|
|
err_R3SVD[i] = 100.0 * abs( S_FSVD[i] - S_R3SVD[i] ) / S_FSVD[i]
|
|
np.savetxt('results_python/h2o_pseudo/SingValues_R3SVD.txt', np.transpose([ np.array(range(rank))+1, S_R3SVD, err_R3SVD ]), fmt=fmt, delimiter=' ')
|
|
# !!!
|
|
nb_oversamp = 10
|
|
tol_SVD = 1e-10
|
|
print('RRR SVD my version:')
|
|
t_beg = time.time()
|
|
U, S_MRSVD, VT = R3SVD_AMMAR(Y, t, delta_t, npow, nb_oversamp, err_thr, maxit, tol_SVD)
|
|
t_end = time.time()
|
|
rank = S_MRSVD.shape[0]
|
|
energy = np.sum( np.square(S_MRSVD) ) / normY**2
|
|
err_SVD = 100. * np.linalg.norm(Y - np.dot(U,np.dot(np.diag(S_MRSVD),VT)), ord='fro') / normY
|
|
print('nb Pow It = {}, rank = {}, energy = {}, error = {} %, CPU time = {}\n'.format(npow, rank, energy, err_SVD, t_end-t_beg))
|
|
# !!!
|
|
err_MRSVD = np.zeros( (rank) )
|
|
for i in range(rank):
|
|
err_MRSVD[i] = 100.0 * abs( S_FSVD[i] - S_MRSVD[i] ) / S_FSVD[i]
|
|
np.savetxt('results_python/h2o_pseudo/SingValues_MRSVD.txt', np.transpose([ np.array(range(rank))+1, S_MRSVD, err_MRSVD ]), fmt=fmt, delimiter=' ')
|
|
# !!!
|
|
trank = rank
|
|
print("Truncated RSVD (pre-fixed rank = {} & oversampling parameter = {}):".format(trank,nb_oversamp))
|
|
t_beg = time.time()
|
|
U, S_RSVD, VT = powit_RSVD(Y, trank, npow, nb_oversamp)
|
|
t_end = time.time()
|
|
rank = S_RSVD.shape[0]
|
|
energy = np.sum( np.square(S_RSVD) ) / normY**2
|
|
err_SVD = 100. * np.linalg.norm( Y - np.dot(U,np.dot(np.diag(S_RSVD),VT)), ord="fro") / normY
|
|
print('nb Pow It = {}, rank = {}, energy = {}, error = {} %, CPU time = {}\n'.format(npow, rank, energy, err_SVD, t_end-t_beg))
|
|
# !!!
|
|
err_RSVD = np.zeros( (rank) )
|
|
for i in range(rank):
|
|
err_RSVD[i] = 100.0 * abs( S_FSVD[i] - S_RSVD[i] ) / S_FSVD[i]
|
|
np.savetxt('results_python/h2o_pseudo/SingValues_RSVD.txt', np.transpose([ np.array(range(rank))+1, S_RSVD, err_RSVD ]), fmt=fmt, delimiter=' ')
|
|
# !!!
|
|
print("Truncated SVD (scipy):")
|
|
t_beg = time.time()
|
|
U, S_TSVD, VT = svds(Y, min(trank, min(Y.shape[0],Y.shape[1])-1 ), which='LM')
|
|
t_end = time.time()
|
|
rank = S_TSVD.shape[0]
|
|
energy = np.sum( np.square(S_TSVD) ) / normY**2
|
|
err_SVD = 100. * np.linalg.norm( Y - np.dot(U, np.dot(np.diag(S_TSVD),VT) ), ord="fro") / normY
|
|
print('rank = {}, energy = {}, error = {} %, CPU time = {}\n'.format(rank, energy, err_SVD, t_end-t_beg))
|
|
# !!!
|
|
err_TSVD = np.zeros( (rank) )
|
|
for i in range(rank-1):
|
|
for j in range(i+1,rank):
|
|
if( S_TSVD[j] > S_TSVD[i]):
|
|
S_TSVD[i], S_TSVD[j] = S_TSVD[j], S_TSVD[i]
|
|
for i in range(rank):
|
|
err_TSVD[i] = 100.0 * abs( S_FSVD[i] - S_TSVD[i] ) / S_FSVD[i]
|
|
np.savetxt('results_python/h2o_pseudo/SingValues_TSVD.txt', np.transpose([ np.array(range(rank))+1, S_TSVD, err_TSVD ]), fmt=fmt, delimiter=' ')
|
|
# !!!
|
|
# !!!
|